JOURNAL OF COMPUTATIONAL PHYSICS 125, 488-497 (1996)

Numerical Modeling of Space-Charge-Limited

Charged-Particle Emission on a Conformal Triangular Mesh
STANLEY HUMPHRIES JR.

Department of Electrical and Computer Engineering, University of New Mexico,

Albuquerque, New Mexico 87131

Received June 9, 1995; revised October 16, 1995

Orbit tracing is a numerical technique to design high-current charged particle guns and
transport systems. Although most available programs use the finite-difference approach
to calculate electric fields, the finite-element method has advantages for accuracy and
versatility. This paper describes an effective algorithm to model spacecharge-limited
emission and a procedure to apply it to variable-resolution conformal triangular meshes
with multiple emission regions. Improved tabulations of space-charge solutions for
sharply curved emission surfaces are given. In a benchmark Pierce diode simulation, the
procedure gives an absolute current prediction within 0.2% of the analytic value and
particle orbits within 0.1/ of the predicted exit angle.

1. INTRODUCTION

High-current electron guns are used in microwave sources, beam welders, materials processing
accelerators, and a range of other applications [1]. Beam-generated electric fields strongly
influence the characteristics of these devices, and numerical calculations are essential for their
design. In most applications, the beam pulse length is much longer than the electron transit time.
In this regime, orbit tracing codes [2] are effective tools to find self-consistent electric fields and
beam current density. Several computer programs exist for electron and ion gun design [3-7].
With the exception of the Demeos code [7], they use finite-difference methods on square or
rectangular meshes to solve the Poisson equation. This paper describes the technique to model
space-charge emission in the finite-element Trak code [8]. The method can accommodate
multiple emission regions with arbitrary source shapes and triangular mesh geometries. Section 2
reviews orbit tracing and finite-element field solutions. The discussion concentrates on
two-dimensional geometries, including the important cases of cylindrically symmetric guns and
slot injectors. The section also describes the assignment of space charge to a finite-element mesh
from model particle traces. Section 3 summarizes a method to calculate particle orbits near an
emission surface. The method gives an accurate representation of local space-charge and avoids
the numerical problem of zero-velocity particles. A one-dimensional model confirms the

1
approach and yields information on accuracy and strategies to ensure convergence. Section 4
discusses application of the method on a two-dimensional finite-element mesh. Section 5
summarizes correction factors to improve the accuracy of space-charge-limited emission
calculations from curved . electrodes. The method makes it easy to model curved and planar
sources with mixed source-limited and space-charge-limited flow.

2. BEAM-GENERATED FIELDS ON A FINITE-ELEMENT MESH

Figure 1 shows a conformal mesh near an electron-emitting surface. The mesh divides the
vacuum space into triangular elements that closely match the boundary surfaces. The size of the
triangles can be adjusted to give fine resolution in regions of strong field variations. The
finite-element expression of the Poisson equation proceeds from the integral form of the
Maxwell equations applied over the triangles that surround each mesh vertex [9-11]. In the linear
approximation, volume quantities like space-charge and dielectric permeability are constant over
each triangle. The numerical form of the Poisson equation relates the potential at each vertex to
the potentials at the neighboring points and the characteristics of the six surrounding triangles.
The following discussions assume a regular mesh with six triangles of area ai adjacent to each
point. Figure 2 defines the mesh geometry near a test point. The numerical equivalent of the
Poisson equation is

Equation (1) states that the vertex potential is a weighted average of the potential values at the
nearest neighbors and the surrounding space-charge. The weighting factors are given by

Figure 2 defines the angles that characterize the triangle geometries. Equation (1) represents a
large set of coupled linear equations. These equations can be solved by iterative relaxation or by
a direct solution using matrix algebra [9].
Space-charge arises from the beam density in chargedparticle gun calculations. The standard
technique to estimate D(x) is to represent a beam with a moderate number of model particles [2].
Because of the laminar behavior of particle distributions under the Vlasov equation, a model
particle can represent the average behavior of many nearby particles in phase space. The

2
FIG. 1. Example of a conformal triangular mesh with variable resolution matched to electrode surfaces.

procedure is to follow the orbits of model particles in the total electric and magnetic fields as if
they were single electrons (or ions), but to assign space-charge along the trajectory as though the
model particle carried the charge of many adjacent particles. The procedure to assign
space-charge to mesh triangles is straightforward. A Runge-Kutta integration [12] of the
equations of the motion with time step At yields a set of closely spaced coordinates x(n At) and
the positions at intermediate time steps, x([n + 2] )t). The indices of the triangle occupied by the
particle at the intermediate time are available from the electric field calculation. Over a time step
in the orbit solution for particle j, the spacecharge density of the triangle is incremented by

In Eq. (2), Ij is the current for model particle j and )V the volume represented by the triangular.
For example, in cylindrical coordinates the volume is )V = 2 Bra )A, where )A is the triangle
area and ra is the radius of the triangle centroid.
The challenge to finding self-consistent electric fields is that the particle orbits are not known
in advance; therefore, the quantity D(x) and the fields are unknown. The resolution in orbit
tracing programs is to use an iterative approach. Initially, the total fields are set equal to the

3
FIG. 2. A point of a finite-element triangular mesh-six neighboring mesh points define six surrounding
triangles. Definitions of quantities for linear approximation of the Poisson equation.

applied field and a set of model particle trajectories are calculated. Space-charge deposition
follows the prescription of Eq. (2). The fields are then recalculated with the addition of D(x). The
corrected fields give modified particle orbits that, in turn, lead to an improved field calculation.
With appropriate charge averaging, the process converges to the correct self-consistent fields,
even for high-intensity beams.
The calculation is more difficult for beams from a spacecharge limited source [13]. Here, the
model particle currents are not known in advance-they depend on the electric field intensity near
the source. The problem is that space-charge-limited emission coincides with zero electric field
at the source surface. In a numerical model, it is impossible to start model particles at this
surface because they will not advance. The solution is to create the particles at a hypothetical
emission surface and to determine the model particle current from a local application of analytic
formulae. Sections 3 and 4 describe the method used in Trak. In contrast to other ray tracing
codes, the method is compatible with arbitrary mesh and source geometries and multiple regions
of emission.

4
FIG. 3. One-dimensional acceleration gap for electrons. Definition of quantities for the Child law
solution and numerical calculation of space-charge-limited emission.

3. ONE-DIMENSIONAL MODEL FOR SPACE-CHARGE LIMITED EMISSION

This section reviews the numerical calculation of space-charge-limited emission in the

one-dimensional gap of Fig. 3 – results can be checked against Child law [14] predictions. The
discussion holds for electrons-it can be easily extended to ions. The gap has width d and an
applied voltage of +Vo. The condition of space-charge-limited flow is that the electron density
approaches a value that reduces the electric field on the cathode surface to zero. For non-
relativistic electrons, the Poisson equation for electrostatic potential is

In Eq. (3), the quantity jo is the current density, a constant over the gap. The solution of the
equation with the following boundary conditions determines the value of jo: N(0) = 0, dN(0)/dx =
0, and N(d) = Vo. Equation (3) can also be written in terms of the dimensionless variables M = N/
Vo and X = x/d as

where

5
The solution of Eq. (4) with M(0) = 0, M’(0) = 0, and M(1) = 1 gives ' = 4/9 = 0.444444,
equivalent to

Equation (6) is the familiar Child law [14]. The electrostatic potential varies as

The electric field is zero at x = 0.0 and has the value

at x = d.
To find a numerical solution for space-charge flow, we apply the Child law over an emission
surface at the cathode of width ds. The surface is much thinner than the gap, ds < d. The potential
of the emission surface relative to the source is Ns(ds). Non-relativistic expressions are valid if ds
is small enough so that eNs n mec2. At the surface electrons have the velocity

Electrons always have nonzero velocity in the region to the right of the emission surface;
therefore, the associated space-charge follows from an integration of the relativistic equations of
motion and the prescription of Eq. (2). Problems arise assigning space charge to mesh elements
to the left of the emission surface. Direct calculations from the analytic Child law expressions
are difficult to extend to a two-dimensional triangular mesh. Writing a program that can reliably
perform volume integrations for arbitrary mesh and emitter surface geometries is challenging.
An alternative is to project model electrons backward to the source from the emission surface
with the velocity magnitude of Eq. (8), assigning space charge according to Eq. (2). This
procedure can lead to density divergences for solutions near the space-charge limit when the
model particle velocity approaches zero.
The Trak program uses a variant of the second method that avoids the problem of density
divergence. To illustrate the method, consider the one-dimensional model of Fig. 3. The goal is
to find a numerical solution of the Poisson equation by tracking a single model particle from the
emission surface. The calculation is initialized with the applied field solution, a uniform electric
field Ex = -Vo/d. The initial potential at the emission surface is

The model particle carries the current I = jA, where

6
and A is a unit area. To find the charge density to the right of the surface, the model electron is
created at the emission surface with velocity

and advanced with uniform time step )t. The problem is how to assign charge to the left of the
emission surface while avoiding a zero velocity model particle. One solution is to take a uniform
density that gives the same value of electric field at the emission surface as the Child law, Eq.
(7). This method has the advantage that a uniform density can be impressed on a two-
dimensional triangular mesh simply by projecting constant-velocity model particles backwards.
This circumvents dealing with the complexities of mesh and emission surface geometries.
Inspection of the solution of the one-dimensional Poisson equation gives the desired density
value Do = 4,oNs/6ds. This density results if the model particle moves backward from the
emission surface to the source with the current density of Eq. (9) and with a constant velocity
equal to two-thirds that of Eq. (10),

This method is easy to implement on a two-dimensional mesh because the logic of particle
charge assignment and tracking is almost identical for the forward and backward orbit
integrations. The only difference is to set electric field values equal to zero during the backward
trace. After several iterations, the emission surface potential approaches

and the model particle current density, I/A, approaches the value of Eq. (6).
One-dimensional finite-element calculations were made to test the procedure and to document
parameter sensitivities. The model used a uniform element size of )x. Backsubstitution [15] was
applied to solve the one-dimensional Poisson equation, and particle orbits were advanced with a
second-order Runge-Kutta routine. The free parameters in the model were NMesh (the number of
elements across the gap), ds/d (the relative distance of the emission surface from the source), and
)t (the time step for the orbit integration). The scaling parameter for the time step was the transit
time across a cell for a particle moving at the exit velocity,

To achieve stable convergence over the range of parameter choices, it was necessary to
introduce two correction factors. The first was a suppression factor, applied on initial iteration
cycles. On the first cycle, the applied electric field at the emission surface was much larger than

7
the final selfconsistent value. Therefore, application of Eq. (9) gives excessive space-charge.
This often caused an instability between cycles-the emission current was alternately enhanced
and depressed near the cathode. One improvement was to use a gentle start, multiplying the
value of Eq. (9) by a graded sequence of numbers less than unity during initial cycles. In
addition, charge averaging between iteration cycles also reduced emission instabilities. Instead
of zeroing the values of Dk for each particle tracking cycle, they were initially set equal to a
fraction of the space charge from the previous cycle.

Here, , is a number between 0.0 and 1.0. During particle tracking, the incremental space-charge
is

A value of , = 1.0 corresponds to no averaging. A value , n 1 gives averaging over several

previous cycles. On the first cycle, , is set equal to 1.0. Generally, a value of , = 0.5 gave stable
onedimensional calculations. A low value is essential for twodimensional calculations with
complex reflex orbits.
Runs were made to check the effect of the mesh size, given by the parameter NMesh, on the
solution accuracy. The emission region occupied 10% of the gap, the space charge averaging
factor was , = 0.5 and the time step was short. The calculations converged in about 15 cycles.
The tabulated error is the difference between the predicted current density and the Child law
value. The error dropped to 0.34% with only 20 mesh boxes. A second test checked the effect of
the emission region width, ds/d. These runs used NMesh = 50 and a short time step. The accuracy
was 0.07% for ds/d = 0.08 and had the acceptable value of 0.43% for an emission extending over
only one mesh element. It is noteworthy that the method converges to the correct value of
current density, even when the emission region fills the entire gap. In the one-dimensional
model, the current density was almost independent of dsld. Nonetheless, it is important to note
that two-dimensional calculations require that ds/d n 1. In this limit the surface conforms to the
source shape and the errors in the transverse electric field resulting from the assumption of
uniform density are correspondingly small.
The final test checked the effect the time step size for NMesh = 50 and ds/d = 0.10. For a value
)t/)to n 1, the model particle took several time steps to cross each mesh element, giving a
smooth distribution of D(x). For the value )t/)to = 20.0, the paricle crossed many mesh elements
in a time step giving an irregular distribution of space-charge. Even for this extreme case the
method converged to a current density value within 0.3% of the Child law value. In summary,
the one-dimensional tests confirm that the constant density procedure gives good accuracy for
)x < d/20, an emission distance of about 0.10 of the acceleration gap and a value of )t such that
particles travel about one mesh unit per time step near the emission surface. In two-dimensional
calculations, these conditions can be easily satisfied on a variable resolution triangular mesh.

8
FIG. 4. Definition of two emission regions on a shaped cathode, showing two sets of ordered source
points, projected emission surface points, and distance spans at the source and emission surfaces.

4. PROCEDURES FOR SPACE-CHARGE-LIMITED EMISSION IN TWO

DIMENSIONS

The space-charge emission method of Section 3 can be applied to two-dimensional calculations

on a finite-element triangular mesh. The main tasks are to identify source points, to set emission
surface points, and to assign effective areas to model particles. The first step is to set emission
surfaces on the electrodes. This task is accomplished during mesh generation by assigning a
special flag to surface points that emit particles. For space-charge deposition, there is little
advantage to choosing the spacing between particles smaller than the local mesh triangle size.
Therefore, the convention in the Trak program is to create a model particle at each of the
emitting source points. The mesh resolution at the source determines the number of model
particles in the beam. The Trak program handles up to 10 separated emission surfaces. The term
set denotes the surface points and model particles that constitute an individual emission surface.
After reading the computational mesh, the initial task of the tracking program is to collect the
source points in each set and to arrange them spatially in order of distance from a reference
point. The user can specify reference points for each set to resolve ordering ambiguities. With
judicious choices of reference points and sets, Trak can model emission from any shaped
surface, including indentations and discontinuities. Figure 4 illustrates the definition of surface
points, sets, and reference points. The figure also shows point ordering. The next task is to

9
compute the distance span of each model particle on the source surface, Ds(i), where i is the
model particle index. These quantities are used to find the effective crosssection areas of the
model particles. For the inside particles of a set, Ds(i) equals the distance between points halfway
to the adjacent particles (Fig. 4). For the two outside particles of a set, there are two options to
set the spans. For a sharp emission edge, the quantity Ds(i) equals the distance from the edge
point halfway to the adjacent point. This convention gives correct total current if the end points
represent the edge of the source. The second option is applied to blend the current density from
two or more adjacent sets. In this case, the distance span equals the full distance to the adjacent
particle of the set.
The third task of the particle tracking program is to determine a set of points at a distance ds
from the source that constitutes the emission surface. Trak performs this operation geometrically.
For each inside point of a set, the program finds a unit vector normal to the local source surface
using the positions of the two adjacent points. The vector is then used to compute the
corresponding emission surface point (Fig. 4). For the end points of a set, the vector calculation
uses the coordinates of the end point and one adjacent point. In general, sources may be concave
or convex-the surface point locations alone do not fully constrain the normal vector. The
procedure in Trak is to move in one direction and then to make a sample electric field
calculation. If the procedure returns an error, the point is inside an electrode or outside the
computational region. In this case, the program tries the opposite direction. This procedure gives
the correct emission surface location for any source shape or orientation-it also signals an error
for source concavities that are too small to enclose the emission surface. Once a set of consistent
surface points is known, Trak computes the emission surface distance spans, De(i), using the
same conventions applied to the source points. These quantities give the effective model particle
areas at the emission surface, Ae(i). For example, the expression in cylindrical coordinates is

where ri is the distance from the emission point to the z axis.

At this point, the program has enough information to apply the space-charge procedure of
Section 3. On each tracking cycle, Trak first calculates values of the electric field and the
potential difference from the source, Ns, at every emission point. For flat or mildly curved
surfaces, substituting the quantities Ns and ds in Eq. (9) gives a good approximation to the
emission surface current density. (Section 5 derives corrections for highly curved surfaces.) The
total current of a model particle is the product of the current density and Ae(i). The program
assigns each electron a constant velocity b(2eNs/me)½ aligned parallel to the local electric field
and tracks it backward to the source, assigning space-charge according to Eq. (9). Next, the
program gives each electron an initial velocity with magnitude (2eNs/me)½ aligned anti-parallel
to the electric field and advances it by the relativistic equations of motion, assigning a charge
along the way. Trak includes a variety of particle orbit termination options-a model particle may
stop if it enters an electrode or dielectric, leaves the solution region, exceeds a maximum time, or
crosses a special interpolation plane. After all the model electron trajectories are complete, the
program updates the electrostatic potential using Eq. (1). The process repeats for several cycles
until the solution converges. The two-dimensional calculations use the gentle-start suppression

10
FIG. 5. Benchmark test of the ideal Pierce diode-structure extends out of page and is symmetric about the
bottom axis: (a) Geometry and electrode potentials (xmin = 0.0 cm, xmax = 2.5 cm, ymin = 0.0 cm, ymax
= 4.0 cm). (b) Variable resolution mesh used for a TRAK simulation. (c) Detailed view of model particle
orbits and equipotential lines for the self-consistent field solution (xmin = 0.0 cm, xmax = 2.5 cm, ymin =
0.0 cm, ymax = 2.5 cm).

11
factors and charge averaging procedures discussed in Section 3.
To demonstrate the accuracy of the methods in a twodimensional calculation, consider the
standard Pierce diode for a sheet electron beam [16]. Figure 5a illustrates the geometry. The
system extends out of the page and has reflection symmetry about the bottom axis. The flat
surface on the left is a proton emitter. The focusing electrode above the emitter inclines at an
angle of 22.5/. This electrode, combined with the shaped cathode and extraction grid on the
right, ensures uniform current density over the source surface and parallel orbits for the exiting
protons. Equation (6) gives the current density with d equal to the distance between the emitter
and the extraction grid and with Vo equal to the anode-cathode voltage. In this example, values of
d = 2.0 cm and Vo = 50 kV give a predicted current density of 0.1528 A/cm2. The Trak
calculation uses the 4500 point mesh of Fig. 5b with variable resolution. Note the resolution
enhancements in the propagation and emission regions. The calculation takes 135 s for 20
tracking cycles on a 90 MHZ Pentium [17] computer. With an emission surface distance of ds =
2 mm, the code prediction of current density is 0.1530 A/cm2. This figure is within 0.2% of the
analytic value. Figure 5c shows the orbits of 61 model particles and equipotential lines for the
self-consistent field. As expected, the equipotential lines are normal to the beam boundary, and
the proton orbits are parallel to within ±0.1/ at the extraction grid. The current density at the
emission surface is uniform to within +0.2%. The accuracy is similar to that reported for the
Demeos code [7]. A comparison calculation was made with Egun [3], a finite difference code
that uses a uniform square mesh. The calculation used 3200 mesh points, 20 tracking cycles, 82
model particles, and the same emission surface distance. The total current prediction was in error
by about 1%. In the final solution, there was a systematic variation of current density and particle
divergence from the axis to the beam outer edge. The current density varied by about 5% and
there was an 0.75/ beam divergence at the edge.

5. CORRECTIONS FOR SHARPLY CURVED ELECTRODES

The application of Eq. (9) to calculate emission surface current density is adequate for many
conventional electron and ion guns, but corrections are necessary for the accurate representation
of emission from sharply curved surfaces. Initially, we shall concentrate on solutions in
rectangular coordinates. Here, surfaces vary in x and y but extend a long distance in z. Figure 6
illustrates a blade emitter. The standard planar Child law (Eq. (9)) applies to the flat regions
where De(i) = Ds(i). Corrections are necessary at the sharply curved edges, where De(i) > Ds(i).
Correction factors can be estimated by assuming that flow in the section of the emission region
corresponding to a model particle approximates space-charge-limited flow in a cylindrical
section. Numerical solutions in cylindrical and spherical coordinates have long been available. A
familiar example is the Langmuir a function for spherical flow [18]. Nonetheless, it is
worthwhile to extend the calculations to determine high-accuracy correction factors that are
better suited to particle tracking codes.
The insert of Fig. 6 illustrates the geometry for diverging flow. If the angular span of an edge
section is much smaller than 1 radian, then the distance spans are related to the inner and outer
radii of the equivalent cylindrical solution by

12
Given an emission surface spacing

and a potential difference of Ns, the goal is to find the current density at the outer radius
(emission surface). It is convenient to compare the solution to the prediction of Eq. (9) for the
same values ds and Ns to define a curvature correction factor, Fc. This number is a function of the
ratio of distance spans,

The values of $i are easily calculated for each model particle during the emission surface setup
procedure. Correction of the current density simply involves multiplying the current density at
each emission point determined from Eq. (9) by Fc( $i).
Consider the diverging flow of electrons in the cylindrical region of Fig. 6. The electrons move
from ri at ground potential to ro at N = Vo, where the difference between the radii equals ds [Eq.
(11)]. If jo is the current density at the emission surface, then the electron space-charge density is
given by

Taking the radial velocity as

the Poisson equation is

Introducing the dimensionless variables M = N/Vo and R = r/ds, we can rewrite the above
equation as

where Ro = ro/ds and r is given by Eq. (5).

Equation (12) holds over the range Ri = ri/ds < R < Ro and satisfies following boundary
conditions:

13
FIG. 6. Cylindrical correction for elements on the curved surface of a blade emitter-structure extends out
of page.

and

The inner and outer radii can be written in terms of $ as

Table I shows the current adjustment factor with accuracy to four decimal places calculated
from 100 shooting cycles. The table encompasses the full practical range. For larger or smaller
values of $, the potential variation near the surface is poorly represented by the assumption of a
cylindrical section. The following series expansion gives Fc accurate to 5 × 10-4 over the range of
the table:

14
FIG. 7. Cylindrical emitter-a figure of revolution about the axis. A spherical correction is applied to
surface element a and a cylindrical correction to element b.

Curvature corrections for systems with cylindrical symmetry are more involved. If the emitter
surface has a center of curvature on the z axis, then space-charge flow in an element is spherical.
This case corresponds to the element marked a in Fig. 7. The spherical Poisson equation has the
form

The above equation has the same boundary conditions as Eq. (12). Equations (14) and (15) relate
the inner and outer dimensionless radii to the parameter $ = De/Ds for diverging and converging
flows. Solution of the above equation gives the spherical correction factor, Fc($), listed in Table
I. The spherical correction factor is related to the Langmuir function [18] by

The quantity Fc is better suited to tabulation and series

approximation because it is close to unity over the range of The table, while the Langmuir
function varies over several orders of magnitude. The following expansion gives a good
approximation to the values of Table I:

15
Spherical corrections do not apply in cylindrical coordinates to cases where the center of
curvature of a surface element does not lie on the z axis. For example, the element marked b in
Fig. 7 is a sharp radius on the outside of a cylindrical emitter. The radial displacement of the
element at the emission surface is almost the same as at the source. Here, it is more accurate to
apply cylindrical corrections. The convention of Trak is to apply cylindrical corrections to all
elements by default – spherical corrections are applied to a set only if requested by the user.
It is easy to model guns with mixed source and space-charge-limited emission. Suppose there
is a limit js on the current density at the source surface. For model particles on each tracking
cycle, Trak computes the predicted current density at the emission surface from the equation

In cylindrical coordinates, the emission surface current density projected to the source is je$i. If
je$i.< js, the emission current density is set equal to ji; otherwise, the current density equals js/$i.
For spherical coordinates, the condition for source limited current is je$i2 > js and the
corresponding emission surface current density is js/$i2.
In conclusion, the finite element approach gives improved accuracy for numerical modeling of
charged particle guns. The main difficulty is dealing with arbitrary mesh geometries. The
problem is challenging for space-charge-limited emission, where there is a singularity at the
particle source. The emission surface procedure described in this paper resolves the problem for
sources of arbitrary shape. To model space charge in the region between the source and emission
surface, the method applies the same techniques used for particle tracking and space-charge
assignment in downstream regions. Good approximations for the space charge and potential
variations can be derived by tracking particles backward from the emission surface to the source
at a specific constant velocity. The method is easily extended to accommodate multiple emission
regions, sharply curved sources, and mixed source-limited and space-charge-limited emission.

ACKNOWLEDGMENTS

This work was partially supported by North Star Research Corporation through Department of
Energy SBIR Grant Number DE-FG0591ER81209. I thank Richard Adler for his suggestions
and encouragement. I am also grateful to the Accelerator Code Group of Los Alamos National
Laboratory for permission to use finite-element routines from the Lattice and Poisson programs
originally developed by Ronald Holsinger.

16
 REFERENCES

1. See, for instance, W. Scharf, Particle Accelerators and Their Uses (Harwood Academic,
Chur, Switzerland, 1986).
2. See, for instance, S. Humphries Jr., Charged Particle Beams (Wiley, New York, 1990),
Section 7.3.
3. W. B. Herrmannsfeldt, Stanford Linear Acc. Center, SLAC-331, 1988 (unpublished).

4. A. C. Paul, Lawrence Berkeley Lab, LBL-13241,1982 (unpublished).

5. D. L. Vogel, Lawrence Berkeley Lab., LBL-18871, 1985 (unpublished).
6. J. E. Boers, Sandia National Labs, SAND 79-1027, 1980 (unpublished).
7. R. True, IEEE Trans. Nucl. Sci. NS-32, 2611 (1985); General Purpose Relativistic Beam
Dynamics Code, in Computational Accelerator Physics, edited by R. Ryne (Am. Inst. of
Phys., New York, 1994), p. 493.
8. S. Humphries Jr., TRAK, in Computational Accelerator Physics, edited by R. Ryne (Am.
Inst. of Phys., New York, 1994), p. 597; Integrated Software System for High-power Beam
Design, in Beams 94: Proceedings, 10th Conf High Power Particle Beams (National
Technical Information Service, NTIS PB95-144317, 1995), p. 568.
9. A. M. Winslow, J. Comput. Phys. 1(2), 149 (1967).
10. Accelerator Code Group, Los Alamos National Laboratory, LAUR-87-126, 1987, Section
B.13.6.2 (unpublished).
11. See, for instance, M. N. O. Sadiku, Numerical Techniques in Electromagnetics (CRC
Press, Boca Raton, FL, 1992), Chap. 6.
12. See, for instance, W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery,
Numerical Recipes in Fortran, 2nd ed. (Cambridge Univ. Press, Cambridge, 1992), Section
16.1.
13. S. Humphries Jr., [2, Section 5.2].
14. C. D. Child, Phys. Rev. 32, 492 (1911).
15. D. Potter, Computational Physics (Wiley, New York, 1973), p. 88.
16. J. R. Pierce, Theory and Design of Electron Beams (Van Nostrand, Princeton, NJ, 1949).
17. Registered trademark of Intel Corporation.
18. I. Langmuir and D. Blodgett, Phys. Rev. 24, 49 (1924).
19. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, [12, Section 17.1].

17
 TABLE 1. Curvature adjustment factors

Connverging beams
Fc Fc
$ Cylinder Sphere
0.300 1.392 1.670
0.325 1.354 1.611
0.350 1.321 1.559
0.375 1.291 1.512
0.400 1.265 1.470
0.425 1.242 1.432
0.450 1.220 1.396
0.475 1.201 1.364
0.500 1.183 1.334
0.525 1.167 1.307
0.550 1.152 1.281
0.575 1.139 1.257
0.600 1.126 1.234
0.625 1.114 1.214
0.650 1.103 1.194
0.675 1.092 1.175
0.700 1.083 1.157
0.725 1.074 1.140
0.750 1.065 1.125
0.775 1.057 1.110
0.800 1.049 1.095
0.825 1.042 1.082
0.850 1.035 1.068
0.875 1.029 1.056
0.900 1.022 1.044
0.925 1.016 1.032
0.950 1.011 1.021
0.975 1.006 1.011
1.000 1.000 1.000

18
Diverging Beams 2.800 0.865 0.680
$ Fc Fc 2.850 0.863 0.676
Cylinder Sphere 2.900 0.862 0.672
1.050 0.991 0.981 2.950 0.861 0.668
1.100 0.982 0.963 3.000 0.860 0.664
1.150 0.974 0.946
1.200 0.966 0.931
1.250 0.960 0.916
1.300 0.953 0.902
1.350 0.947 0.889
1.400 0.942 0.877
1.450 0.936 0.865
1.500 0.931 0.854
1.550 0.927 0.843
1.600 0.923 0.833
1.650 0.918 0.824
1.700 0.915 0.815
1.750 0.911 0.806
1.800 0.908 0.798
1.850 0.904 0.789
1.900 0.901 0.782
1.950 0.898 0.774
2.000 0.896 0.767
2.050 0.893 0.760
2.100 0.890 0.754
2.150 0.888 0.747
2.200 0.886 0.741
2.250 0.883 0.735
2.300 0.881 0.729
2.350 0.879 0.724
2.400 0.878 0.718
2.450 0.876 0.713
2.500 0.874 0.708
2.550 0.872 0.703
2.600 0.871 0.698
2.650 0.869 0.694
2.700 0.868 0.689
2.750 0.866 0.684

19